xref: /linux/drivers/md/dm-log-userspace-base.c (revision d4880868670198df321627a949e7b7f2d76cf54e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2006-2009 Red Hat, Inc.
4  *
5  * This file is released under the LGPL.
6  */
7 
8 #include <linux/bio.h>
9 #include <linux/slab.h>
10 #include <linux/jiffies.h>
11 #include <linux/dm-dirty-log.h>
12 #include <linux/device-mapper.h>
13 #include <linux/dm-log-userspace.h>
14 #include <linux/module.h>
15 #include <linux/workqueue.h>
16 
17 #include "dm-log-userspace-transfer.h"
18 
19 #define DM_LOG_USERSPACE_VSN "1.3.0"
20 
21 #define FLUSH_ENTRY_POOL_SIZE 16
22 
23 struct dm_dirty_log_flush_entry {
24 	int type;
25 	region_t region;
26 	struct list_head list;
27 };
28 
29 /*
30  * This limit on the number of mark and clear request is, to a degree,
31  * arbitrary.  However, there is some basis for the choice in the limits
32  * imposed on the size of data payload by dm-log-userspace-transfer.c:
33  * dm_consult_userspace().
34  */
35 #define MAX_FLUSH_GROUP_COUNT 32
36 
37 struct log_c {
38 	struct dm_target *ti;
39 	struct dm_dev *log_dev;
40 
41 	char *usr_argv_str;
42 	uint32_t usr_argc;
43 
44 	uint32_t region_size;
45 	region_t region_count;
46 	uint64_t luid;
47 	char uuid[DM_UUID_LEN];
48 
49 	/*
50 	 * Mark and clear requests are held until a flush is issued
51 	 * so that we can group, and thereby limit, the amount of
52 	 * network traffic between kernel and userspace.  The 'flush_lock'
53 	 * is used to protect these lists.
54 	 */
55 	spinlock_t flush_lock;
56 	struct list_head mark_list;
57 	struct list_head clear_list;
58 
59 	/*
60 	 * in_sync_hint gets set when doing is_remote_recovering.  It
61 	 * represents the first region that needs recovery.  IOW, the
62 	 * first zero bit of sync_bits.  This can be useful for to limit
63 	 * traffic for calls like is_remote_recovering and get_resync_work,
64 	 * but be take care in its use for anything else.
65 	 */
66 	uint64_t in_sync_hint;
67 
68 	/*
69 	 * Workqueue for flush of clear region requests.
70 	 */
71 	struct workqueue_struct *dmlog_wq;
72 	struct delayed_work flush_log_work;
73 	atomic_t sched_flush;
74 
75 	/*
76 	 * Combine userspace flush and mark requests for efficiency.
77 	 */
78 	uint32_t integrated_flush;
79 
80 	mempool_t flush_entry_pool;
81 };
82 
83 static struct kmem_cache *_flush_entry_cache;
84 
85 static int userspace_do_request(struct log_c *lc, const char *uuid,
86 				int request_type, char *data, size_t data_size,
87 				char *rdata, size_t *rdata_size)
88 {
89 	int r;
90 
91 	/*
92 	 * If the server isn't there, -ESRCH is returned,
93 	 * and we must keep trying until the server is
94 	 * restored.
95 	 */
96 retry:
97 	r = dm_consult_userspace(uuid, lc->luid, request_type, data,
98 				 data_size, rdata, rdata_size);
99 
100 	if (r != -ESRCH)
101 		return r;
102 
103 	DMERR(" Userspace log server not found.");
104 	while (1) {
105 		set_current_state(TASK_INTERRUPTIBLE);
106 		schedule_timeout(2*HZ);
107 		DMWARN("Attempting to contact userspace log server...");
108 		r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_CTR,
109 					 lc->usr_argv_str,
110 					 strlen(lc->usr_argv_str) + 1,
111 					 NULL, NULL);
112 		if (!r)
113 			break;
114 	}
115 	DMINFO("Reconnected to userspace log server... DM_ULOG_CTR complete");
116 	r = dm_consult_userspace(uuid, lc->luid, DM_ULOG_RESUME, NULL,
117 				 0, NULL, NULL);
118 	if (!r)
119 		goto retry;
120 
121 	DMERR("Error trying to resume userspace log: %d", r);
122 
123 	return -ESRCH;
124 }
125 
126 static int build_constructor_string(struct dm_target *ti,
127 				    unsigned int argc, char **argv,
128 				    char **ctr_str)
129 {
130 	int i, str_size;
131 	char *str = NULL;
132 
133 	*ctr_str = NULL;
134 
135 	/*
136 	 * Determine overall size of the string.
137 	 */
138 	for (i = 0, str_size = 0; i < argc; i++)
139 		str_size += strlen(argv[i]) + 1; /* +1 for space between args */
140 
141 	str_size += 20; /* Max number of chars in a printed u64 number */
142 
143 	str = kzalloc(str_size, GFP_KERNEL);
144 	if (!str) {
145 		DMWARN("Unable to allocate memory for constructor string");
146 		return -ENOMEM;
147 	}
148 
149 	str_size = sprintf(str, "%llu", (unsigned long long)ti->len);
150 	for (i = 0; i < argc; i++)
151 		str_size += sprintf(str + str_size, " %s", argv[i]);
152 
153 	*ctr_str = str;
154 	return str_size;
155 }
156 
157 static void do_flush(struct work_struct *work)
158 {
159 	int r;
160 	struct log_c *lc = container_of(work, struct log_c, flush_log_work.work);
161 
162 	atomic_set(&lc->sched_flush, 0);
163 
164 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH, NULL, 0, NULL, NULL);
165 
166 	if (r)
167 		dm_table_event(lc->ti->table);
168 }
169 
170 /*
171  * userspace_ctr
172  *
173  * argv contains:
174  *	<UUID> [integrated_flush] <other args>
175  * Where 'other args' are the userspace implementation-specific log
176  * arguments.
177  *
178  * Example:
179  *	<UUID> [integrated_flush] clustered-disk <arg count> <log dev>
180  *	<region_size> [[no]sync]
181  *
182  * This module strips off the <UUID> and uses it for identification
183  * purposes when communicating with userspace about a log.
184  *
185  * If integrated_flush is defined, the kernel combines flush
186  * and mark requests.
187  *
188  * The rest of the line, beginning with 'clustered-disk', is passed
189  * to the userspace ctr function.
190  */
191 static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
192 			 unsigned int argc, char **argv)
193 {
194 	int r = 0;
195 	int str_size;
196 	char *ctr_str = NULL;
197 	struct log_c *lc = NULL;
198 	uint64_t rdata;
199 	size_t rdata_size = sizeof(rdata);
200 	char *devices_rdata = NULL;
201 	size_t devices_rdata_size = DM_NAME_LEN;
202 
203 	if (argc < 3) {
204 		DMWARN("Too few arguments to userspace dirty log");
205 		return -EINVAL;
206 	}
207 
208 	lc = kzalloc(sizeof(*lc), GFP_KERNEL);
209 	if (!lc) {
210 		DMWARN("Unable to allocate userspace log context.");
211 		return -ENOMEM;
212 	}
213 
214 	/* The ptr value is sufficient for local unique id */
215 	lc->luid = (unsigned long)lc;
216 
217 	lc->ti = ti;
218 
219 	if (strlen(argv[0]) > (DM_UUID_LEN - 1)) {
220 		DMWARN("UUID argument too long.");
221 		kfree(lc);
222 		return -EINVAL;
223 	}
224 
225 	lc->usr_argc = argc;
226 
227 	strscpy(lc->uuid, argv[0], sizeof(lc->uuid));
228 	argc--;
229 	argv++;
230 	spin_lock_init(&lc->flush_lock);
231 	INIT_LIST_HEAD(&lc->mark_list);
232 	INIT_LIST_HEAD(&lc->clear_list);
233 
234 	if (!strcasecmp(argv[0], "integrated_flush")) {
235 		lc->integrated_flush = 1;
236 		argc--;
237 		argv++;
238 	}
239 
240 	str_size = build_constructor_string(ti, argc, argv, &ctr_str);
241 	if (str_size < 0) {
242 		kfree(lc);
243 		return str_size;
244 	}
245 
246 	devices_rdata = kzalloc(devices_rdata_size, GFP_KERNEL);
247 	if (!devices_rdata) {
248 		DMERR("Failed to allocate memory for device information");
249 		r = -ENOMEM;
250 		goto out;
251 	}
252 
253 	r = mempool_init_slab_pool(&lc->flush_entry_pool, FLUSH_ENTRY_POOL_SIZE,
254 				   _flush_entry_cache);
255 	if (r) {
256 		DMERR("Failed to create flush_entry_pool");
257 		goto out;
258 	}
259 
260 	/*
261 	 * Send table string and get back any opened device.
262 	 */
263 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_CTR,
264 				 ctr_str, str_size,
265 				 devices_rdata, &devices_rdata_size);
266 
267 	if (r < 0) {
268 		if (r == -ESRCH)
269 			DMERR("Userspace log server not found");
270 		else
271 			DMERR("Userspace log server failed to create log");
272 		goto out;
273 	}
274 
275 	/* Since the region size does not change, get it now */
276 	rdata_size = sizeof(rdata);
277 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_GET_REGION_SIZE,
278 				 NULL, 0, (char *)&rdata, &rdata_size);
279 
280 	if (r) {
281 		DMERR("Failed to get region size of dirty log");
282 		goto out;
283 	}
284 
285 	lc->region_size = (uint32_t)rdata;
286 	lc->region_count = dm_sector_div_up(ti->len, lc->region_size);
287 
288 	if (devices_rdata_size) {
289 		if (devices_rdata[devices_rdata_size - 1] != '\0') {
290 			DMERR("DM_ULOG_CTR device return string not properly terminated");
291 			r = -EINVAL;
292 			goto out;
293 		}
294 		r = dm_get_device(ti, devices_rdata,
295 				  dm_table_get_mode(ti->table), &lc->log_dev);
296 		if (r)
297 			DMERR("Failed to register %s with device-mapper",
298 			      devices_rdata);
299 	}
300 
301 	if (lc->integrated_flush) {
302 		lc->dmlog_wq = alloc_workqueue("dmlogd",
303 					       WQ_MEM_RECLAIM | WQ_PERCPU, 0);
304 		if (!lc->dmlog_wq) {
305 			DMERR("couldn't start dmlogd");
306 			r = -ENOMEM;
307 			goto out;
308 		}
309 
310 		INIT_DELAYED_WORK(&lc->flush_log_work, do_flush);
311 		atomic_set(&lc->sched_flush, 0);
312 	}
313 
314 out:
315 	kfree(devices_rdata);
316 	if (r) {
317 		mempool_exit(&lc->flush_entry_pool);
318 		kfree(lc);
319 		kfree(ctr_str);
320 	} else {
321 		lc->usr_argv_str = ctr_str;
322 		log->context = lc;
323 	}
324 
325 	return r;
326 }
327 
328 static void userspace_dtr(struct dm_dirty_log *log)
329 {
330 	struct log_c *lc = log->context;
331 
332 	if (lc->integrated_flush) {
333 		/* flush workqueue */
334 		if (atomic_read(&lc->sched_flush))
335 			flush_delayed_work(&lc->flush_log_work);
336 
337 		destroy_workqueue(lc->dmlog_wq);
338 	}
339 
340 	(void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
341 				    NULL, 0, NULL, NULL);
342 
343 	if (lc->log_dev)
344 		dm_put_device(lc->ti, lc->log_dev);
345 
346 	mempool_exit(&lc->flush_entry_pool);
347 
348 	kfree(lc->usr_argv_str);
349 	kfree(lc);
350 }
351 
352 static int userspace_presuspend(struct dm_dirty_log *log)
353 {
354 	int r;
355 	struct log_c *lc = log->context;
356 
357 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_PRESUSPEND,
358 				 NULL, 0, NULL, NULL);
359 
360 	return r;
361 }
362 
363 static int userspace_postsuspend(struct dm_dirty_log *log)
364 {
365 	int r;
366 	struct log_c *lc = log->context;
367 
368 	/*
369 	 * Run planned flush earlier.
370 	 */
371 	if (lc->integrated_flush && atomic_read(&lc->sched_flush))
372 		flush_delayed_work(&lc->flush_log_work);
373 
374 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND,
375 				 NULL, 0, NULL, NULL);
376 
377 	return r;
378 }
379 
380 static int userspace_resume(struct dm_dirty_log *log)
381 {
382 	int r;
383 	struct log_c *lc = log->context;
384 
385 	lc->in_sync_hint = 0;
386 	r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_RESUME,
387 				 NULL, 0, NULL, NULL);
388 
389 	return r;
390 }
391 
392 static uint32_t userspace_get_region_size(struct dm_dirty_log *log)
393 {
394 	struct log_c *lc = log->context;
395 
396 	return lc->region_size;
397 }
398 
399 /*
400  * userspace_is_clean
401  *
402  * Check whether a region is clean.  If there is any sort of
403  * failure when consulting the server, we return not clean.
404  *
405  * Returns: 1 if clean, 0 otherwise
406  */
407 static int userspace_is_clean(struct dm_dirty_log *log, region_t region)
408 {
409 	int r;
410 	uint64_t region64 = (uint64_t)region;
411 	int64_t is_clean;
412 	size_t rdata_size;
413 	struct log_c *lc = log->context;
414 
415 	rdata_size = sizeof(is_clean);
416 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_CLEAN,
417 				 (char *)&region64, sizeof(region64),
418 				 (char *)&is_clean, &rdata_size);
419 
420 	return (r) ? 0 : (int)is_clean;
421 }
422 
423 /*
424  * userspace_in_sync
425  *
426  * Check if the region is in-sync.  If there is any sort
427  * of failure when consulting the server, we assume that
428  * the region is not in sync.
429  *
430  * If 'can_block' is set, return immediately
431  *
432  * Returns: 1 if in-sync, 0 if not-in-sync, -EWOULDBLOCK
433  */
434 static int userspace_in_sync(struct dm_dirty_log *log, region_t region,
435 			     int can_block)
436 {
437 	int r;
438 	uint64_t region64 = region;
439 	int64_t in_sync;
440 	size_t rdata_size;
441 	struct log_c *lc = log->context;
442 
443 	/*
444 	 * We can never respond directly - even if in_sync_hint is
445 	 * set.  This is because another machine could see a device
446 	 * failure and mark the region out-of-sync.  If we don't go
447 	 * to userspace to ask, we might think the region is in-sync
448 	 * and allow a read to pick up data that is stale.  (This is
449 	 * very unlikely if a device actually fails; but it is very
450 	 * likely if a connection to one device from one machine fails.)
451 	 *
452 	 * There still might be a problem if the mirror caches the region
453 	 * state as in-sync... but then this call would not be made.  So,
454 	 * that is a mirror problem.
455 	 */
456 	if (!can_block)
457 		return -EWOULDBLOCK;
458 
459 	rdata_size = sizeof(in_sync);
460 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_IN_SYNC,
461 				 (char *)&region64, sizeof(region64),
462 				 (char *)&in_sync, &rdata_size);
463 	return (r) ? 0 : (int)in_sync;
464 }
465 
466 static int flush_one_by_one(struct log_c *lc, struct list_head *flush_list)
467 {
468 	int r = 0;
469 	struct dm_dirty_log_flush_entry *fe;
470 
471 	list_for_each_entry(fe, flush_list, list) {
472 		r = userspace_do_request(lc, lc->uuid, fe->type,
473 					 (char *)&fe->region,
474 					 sizeof(fe->region),
475 					 NULL, NULL);
476 		if (r)
477 			break;
478 	}
479 
480 	return r;
481 }
482 
483 static int flush_by_group(struct log_c *lc, struct list_head *flush_list,
484 			  int flush_with_payload)
485 {
486 	int r = 0;
487 	int count;
488 	uint32_t type = 0;
489 	struct dm_dirty_log_flush_entry *fe, *tmp_fe;
490 	LIST_HEAD(tmp_list);
491 	uint64_t group[MAX_FLUSH_GROUP_COUNT];
492 
493 	/*
494 	 * Group process the requests
495 	 */
496 	while (!list_empty(flush_list)) {
497 		count = 0;
498 
499 		list_for_each_entry_safe(fe, tmp_fe, flush_list, list) {
500 			group[count] = fe->region;
501 			count++;
502 
503 			list_move(&fe->list, &tmp_list);
504 
505 			type = fe->type;
506 			if (count >= MAX_FLUSH_GROUP_COUNT)
507 				break;
508 		}
509 
510 		if (flush_with_payload) {
511 			r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
512 						 (char *)(group),
513 						 count * sizeof(uint64_t),
514 						 NULL, NULL);
515 			/*
516 			 * Integrated flush failed.
517 			 */
518 			if (r)
519 				break;
520 		} else {
521 			r = userspace_do_request(lc, lc->uuid, type,
522 						 (char *)(group),
523 						 count * sizeof(uint64_t),
524 						 NULL, NULL);
525 			if (r) {
526 				/*
527 				 * Group send failed.  Attempt one-by-one.
528 				 */
529 				list_splice_init(&tmp_list, flush_list);
530 				r = flush_one_by_one(lc, flush_list);
531 				break;
532 			}
533 		}
534 	}
535 
536 	/*
537 	 * Must collect flush_entrys that were successfully processed
538 	 * as a group so that they will be free'd by the caller.
539 	 */
540 	list_splice_init(&tmp_list, flush_list);
541 
542 	return r;
543 }
544 
545 /*
546  * userspace_flush
547  *
548  * This function is ok to block.
549  * The flush happens in two stages.  First, it sends all
550  * clear/mark requests that are on the list.  Then it
551  * tells the server to commit them.  This gives the
552  * server a chance to optimise the commit, instead of
553  * doing it for every request.
554  *
555  * Additionally, we could implement another thread that
556  * sends the requests up to the server - reducing the
557  * load on flush.  Then the flush would have less in
558  * the list and be responsible for the finishing commit.
559  *
560  * Returns: 0 on success, < 0 on failure
561  */
562 static int userspace_flush(struct dm_dirty_log *log)
563 {
564 	int r = 0;
565 	unsigned long flags;
566 	struct log_c *lc = log->context;
567 	LIST_HEAD(mark_list);
568 	LIST_HEAD(clear_list);
569 	int mark_list_is_empty;
570 	int clear_list_is_empty;
571 	struct dm_dirty_log_flush_entry *fe, *tmp_fe;
572 	mempool_t *flush_entry_pool = &lc->flush_entry_pool;
573 
574 	spin_lock_irqsave(&lc->flush_lock, flags);
575 	list_splice_init(&lc->mark_list, &mark_list);
576 	list_splice_init(&lc->clear_list, &clear_list);
577 	spin_unlock_irqrestore(&lc->flush_lock, flags);
578 
579 	mark_list_is_empty = list_empty(&mark_list);
580 	clear_list_is_empty = list_empty(&clear_list);
581 
582 	if (mark_list_is_empty && clear_list_is_empty)
583 		return 0;
584 
585 	r = flush_by_group(lc, &clear_list, 0);
586 	if (r)
587 		goto out;
588 
589 	if (!lc->integrated_flush) {
590 		r = flush_by_group(lc, &mark_list, 0);
591 		if (r)
592 			goto out;
593 		r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
594 					 NULL, 0, NULL, NULL);
595 		goto out;
596 	}
597 
598 	/*
599 	 * Send integrated flush request with mark_list as payload.
600 	 */
601 	r = flush_by_group(lc, &mark_list, 1);
602 	if (r)
603 		goto out;
604 
605 	if (mark_list_is_empty && !atomic_read(&lc->sched_flush)) {
606 		/*
607 		 * When there are only clear region requests,
608 		 * we schedule a flush in the future.
609 		 */
610 		queue_delayed_work(lc->dmlog_wq, &lc->flush_log_work, 3 * HZ);
611 		atomic_set(&lc->sched_flush, 1);
612 	} else {
613 		/*
614 		 * Cancel pending flush because we
615 		 * have already flushed in mark_region.
616 		 */
617 		cancel_delayed_work(&lc->flush_log_work);
618 		atomic_set(&lc->sched_flush, 0);
619 	}
620 
621 out:
622 	/*
623 	 * We can safely remove these entries, even after failure.
624 	 * Calling code will receive an error and will know that
625 	 * the log facility has failed.
626 	 */
627 	list_for_each_entry_safe(fe, tmp_fe, &mark_list, list) {
628 		list_del(&fe->list);
629 		mempool_free(fe, flush_entry_pool);
630 	}
631 	list_for_each_entry_safe(fe, tmp_fe, &clear_list, list) {
632 		list_del(&fe->list);
633 		mempool_free(fe, flush_entry_pool);
634 	}
635 
636 	if (r)
637 		dm_table_event(lc->ti->table);
638 
639 	return r;
640 }
641 
642 /*
643  * userspace_mark_region
644  *
645  * This function should avoid blocking unless absolutely required.
646  * (Memory allocation is valid for blocking.)
647  */
648 static void userspace_mark_region(struct dm_dirty_log *log, region_t region)
649 {
650 	unsigned long flags;
651 	struct log_c *lc = log->context;
652 	struct dm_dirty_log_flush_entry *fe;
653 
654 	/* Wait for an allocation, but _never_ fail */
655 	fe = mempool_alloc(&lc->flush_entry_pool, GFP_NOIO);
656 	BUG_ON(!fe);
657 
658 	spin_lock_irqsave(&lc->flush_lock, flags);
659 	fe->type = DM_ULOG_MARK_REGION;
660 	fe->region = region;
661 	list_add(&fe->list, &lc->mark_list);
662 	spin_unlock_irqrestore(&lc->flush_lock, flags);
663 }
664 
665 /*
666  * userspace_clear_region
667  *
668  * This function must not block.
669  * So, the alloc can't block.  In the worst case, it is ok to
670  * fail.  It would simply mean we can't clear the region.
671  * Does nothing to current sync context, but does mean
672  * the region will be re-sync'ed on a reload of the mirror
673  * even though it is in-sync.
674  */
675 static void userspace_clear_region(struct dm_dirty_log *log, region_t region)
676 {
677 	unsigned long flags;
678 	struct log_c *lc = log->context;
679 	struct dm_dirty_log_flush_entry *fe;
680 
681 	/*
682 	 * If we fail to allocate, we skip the clearing of
683 	 * the region.  This doesn't hurt us in any way, except
684 	 * to cause the region to be resync'ed when the
685 	 * device is activated next time.
686 	 */
687 	fe = mempool_alloc(&lc->flush_entry_pool, GFP_ATOMIC);
688 	if (!fe) {
689 		DMERR("Failed to allocate memory to clear region.");
690 		return;
691 	}
692 
693 	spin_lock_irqsave(&lc->flush_lock, flags);
694 	fe->type = DM_ULOG_CLEAR_REGION;
695 	fe->region = region;
696 	list_add(&fe->list, &lc->clear_list);
697 	spin_unlock_irqrestore(&lc->flush_lock, flags);
698 }
699 
700 /*
701  * userspace_get_resync_work
702  *
703  * Get a region that needs recovery.  It is valid to return
704  * an error for this function.
705  *
706  * Returns: 1 if region filled, 0 if no work, <0 on error
707  */
708 static int userspace_get_resync_work(struct dm_dirty_log *log, region_t *region)
709 {
710 	int r;
711 	size_t rdata_size;
712 	struct log_c *lc = log->context;
713 	struct {
714 		int64_t i; /* 64-bit for mix arch compatibility */
715 		region_t r;
716 	} pkg;
717 
718 	if (lc->in_sync_hint >= lc->region_count)
719 		return 0;
720 
721 	rdata_size = sizeof(pkg);
722 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_RESYNC_WORK,
723 				 NULL, 0, (char *)&pkg, &rdata_size);
724 
725 	*region = pkg.r;
726 	return (r) ? r : (int)pkg.i;
727 }
728 
729 /*
730  * userspace_set_region_sync
731  *
732  * Set the sync status of a given region.  This function
733  * must not fail.
734  */
735 static void userspace_set_region_sync(struct dm_dirty_log *log,
736 				      region_t region, int in_sync)
737 {
738 	struct log_c *lc = log->context;
739 	struct {
740 		region_t r;
741 		int64_t i;
742 	} pkg;
743 
744 	pkg.r = region;
745 	pkg.i = (int64_t)in_sync;
746 
747 	(void) userspace_do_request(lc, lc->uuid, DM_ULOG_SET_REGION_SYNC,
748 				    (char *)&pkg, sizeof(pkg), NULL, NULL);
749 
750 	/*
751 	 * It would be nice to be able to report failures.
752 	 * However, it is easy enough to detect and resolve.
753 	 */
754 }
755 
756 /*
757  * userspace_get_sync_count
758  *
759  * If there is any sort of failure when consulting the server,
760  * we assume that the sync count is zero.
761  *
762  * Returns: sync count on success, 0 on failure
763  */
764 static region_t userspace_get_sync_count(struct dm_dirty_log *log)
765 {
766 	int r;
767 	size_t rdata_size;
768 	uint64_t sync_count;
769 	struct log_c *lc = log->context;
770 
771 	rdata_size = sizeof(sync_count);
772 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_GET_SYNC_COUNT,
773 				 NULL, 0, (char *)&sync_count, &rdata_size);
774 
775 	if (r)
776 		return 0;
777 
778 	if (sync_count >= lc->region_count)
779 		lc->in_sync_hint = lc->region_count;
780 
781 	return (region_t)sync_count;
782 }
783 
784 /*
785  * userspace_status
786  *
787  * Returns: amount of space consumed
788  */
789 static int userspace_status(struct dm_dirty_log *log, status_type_t status_type,
790 			    char *result, unsigned int maxlen)
791 {
792 	int r = 0;
793 	char *table_args;
794 	size_t sz = (size_t)maxlen;
795 	struct log_c *lc = log->context;
796 
797 	switch (status_type) {
798 	case STATUSTYPE_INFO:
799 		r = userspace_do_request(lc, lc->uuid, DM_ULOG_STATUS_INFO,
800 					 NULL, 0, result, &sz);
801 
802 		if (r) {
803 			sz = 0;
804 			DMEMIT("%s 1 COM_FAILURE", log->type->name);
805 		}
806 		break;
807 	case STATUSTYPE_TABLE:
808 		sz = 0;
809 		table_args = strchr(lc->usr_argv_str, ' ');
810 		BUG_ON(!table_args); /* There will always be a ' ' */
811 		table_args++;
812 
813 		DMEMIT("%s %u %s ", log->type->name, lc->usr_argc, lc->uuid);
814 		if (lc->integrated_flush)
815 			DMEMIT("integrated_flush ");
816 		DMEMIT("%s ", table_args);
817 		break;
818 	case STATUSTYPE_IMA:
819 		*result = '\0';
820 		break;
821 	}
822 	return (r) ? 0 : (int)sz;
823 }
824 
825 /*
826  * userspace_is_remote_recovering
827  *
828  * Returns: 1 if region recovering, 0 otherwise
829  */
830 static int userspace_is_remote_recovering(struct dm_dirty_log *log,
831 					  region_t region)
832 {
833 	int r;
834 	uint64_t region64 = region;
835 	struct log_c *lc = log->context;
836 	static unsigned long limit;
837 	struct {
838 		int64_t is_recovering;
839 		uint64_t in_sync_hint;
840 	} pkg;
841 	size_t rdata_size = sizeof(pkg);
842 
843 	/*
844 	 * Once the mirror has been reported to be in-sync,
845 	 * it will never again ask for recovery work.  So,
846 	 * we can safely say there is not a remote machine
847 	 * recovering if the device is in-sync.  (in_sync_hint
848 	 * must be reset at resume time.)
849 	 */
850 	if (region < lc->in_sync_hint)
851 		return 0;
852 	else if (time_after(limit, jiffies))
853 		return 1;
854 
855 	limit = jiffies + (HZ / 4);
856 	r = userspace_do_request(lc, lc->uuid, DM_ULOG_IS_REMOTE_RECOVERING,
857 				 (char *)&region64, sizeof(region64),
858 				 (char *)&pkg, &rdata_size);
859 	if (r)
860 		return 1;
861 
862 	lc->in_sync_hint = pkg.in_sync_hint;
863 
864 	return (int)pkg.is_recovering;
865 }
866 
867 static struct dm_dirty_log_type _userspace_type = {
868 	.name = "userspace",
869 	.module = THIS_MODULE,
870 	.ctr = userspace_ctr,
871 	.dtr = userspace_dtr,
872 	.presuspend = userspace_presuspend,
873 	.postsuspend = userspace_postsuspend,
874 	.resume = userspace_resume,
875 	.get_region_size = userspace_get_region_size,
876 	.is_clean = userspace_is_clean,
877 	.in_sync = userspace_in_sync,
878 	.flush = userspace_flush,
879 	.mark_region = userspace_mark_region,
880 	.clear_region = userspace_clear_region,
881 	.get_resync_work = userspace_get_resync_work,
882 	.set_region_sync = userspace_set_region_sync,
883 	.get_sync_count = userspace_get_sync_count,
884 	.status = userspace_status,
885 	.is_remote_recovering = userspace_is_remote_recovering,
886 };
887 
888 static int __init userspace_dirty_log_init(void)
889 {
890 	int r = 0;
891 
892 	_flush_entry_cache = KMEM_CACHE(dm_dirty_log_flush_entry, 0);
893 	if (!_flush_entry_cache) {
894 		DMWARN("Unable to create flush_entry_cache: No memory.");
895 		return -ENOMEM;
896 	}
897 
898 	r = dm_ulog_tfr_init();
899 	if (r) {
900 		DMWARN("Unable to initialize userspace log communications");
901 		kmem_cache_destroy(_flush_entry_cache);
902 		return r;
903 	}
904 
905 	r = dm_dirty_log_type_register(&_userspace_type);
906 	if (r) {
907 		DMWARN("Couldn't register userspace dirty log type");
908 		dm_ulog_tfr_exit();
909 		kmem_cache_destroy(_flush_entry_cache);
910 		return r;
911 	}
912 
913 	DMINFO("version " DM_LOG_USERSPACE_VSN " loaded");
914 	return 0;
915 }
916 
917 static void __exit userspace_dirty_log_exit(void)
918 {
919 	dm_dirty_log_type_unregister(&_userspace_type);
920 	dm_ulog_tfr_exit();
921 	kmem_cache_destroy(_flush_entry_cache);
922 
923 	DMINFO("version " DM_LOG_USERSPACE_VSN " unloaded");
924 }
925 
926 module_init(userspace_dirty_log_init);
927 module_exit(userspace_dirty_log_exit);
928 
929 MODULE_DESCRIPTION(DM_NAME " userspace dirty log link");
930 MODULE_AUTHOR("Jonathan Brassow <dm-devel@lists.linux.dev>");
931 MODULE_LICENSE("GPL");
932